Metadata Quality and Discoverability in the NCCU CoDE Open Data Hub: A FAIR- and FGDC CSDGM-Informed Assessment

Abstract

University geospatial hubs support research dissemination, teaching, public data access, and community engagement, but their value depends on whether resources can be found, accessed, interpreted, integrated, and reused. This study evaluates 747 publicly listed items in North Carolina Central University’s CoDE Open Data Hub using FAIR principles as an interpretive framework and the Federal Geographic Data Committee Content Standard for Digital Geospatial Metadata (FGDC CSDGM) as the domain-specific measurement basis. Nineteen criteria were scored on explicit criterion-specific 0–2 rules using live ArcGIS item properties, formal metadata XML, and available layer properties. Discoverability was evaluated for a proportionally stratified sample of 153 items through exact-title, keyword, location, item-type, category-browsing, and click-depth tests. The equal-criterion benchmark averaged 28.57 of 38 (75.19%); equal FAIR-dimension weighting produced a mean of 73.26%, and 95.85% of items retained the same descriptive performance band. Independent rescoring of 50 stratified items produced 76.74% exact criterion-level agreement, a linear weighted Cohen’s kappa of 0.587, and an absolute-agreement ICC of 0.715 for total scores. FAIR-aligned scores were highest for Findability (87.25%) and Accessibility (85.96%), followed by Reusability (67.26%) and Interoperability (52.59%). Exact-title and item-type searches retrieved all sampled items, while keyword search was successful for 88.24%. The results distinguish public availability from technical reuse readiness and identify spatial reference, attribute definitions, lineage, limitations, category structure, and navigation depth as priorities. The performance bands are study-specific descriptive summaries rather than FGDC compliance levels or FAIR certification. The study provides an evidence-preserving, type-aware method for applying FAIR principles alongside geospatial metadata standards in heterogeneous ArcGIS Hub catalogs.

1. Introduction

Web-based geospatial platforms now serve as core components of spatial data infrastructures, allowing institutions to publish data, maps, applications, documents, and analytical products through searchable public catalogs. Within universities, these platforms also support teaching, research dissemination, student project publication, and community-engaged scholarship. Their usefulness, however, depends not only on the volume of available content but also on whether users can find, interpret, evaluate, and reuse individual resources [1,2,3,4].

Metadata is the principal mechanism through which these functions are supported. Clear titles, descriptions, keywords, geographic coverage, dates, source information, lineage, spatial reference, attribute definitions, distribution options, and use constraints determine whether a resource can be found, accessed, integrated, and reused. These concerns align with the FAIR principles, which frame effective data stewardship in terms of Findability, Accessibility, Interoperability, and Reusability [5]. Research on metadata quality has similarly emphasized completeness, accuracy, provenance, consistency, timeliness, and accessibility as conditions for effective discovery and reuse [6,7,8]. Large-scale studies of open data portals show that inconsistent or incomplete metadata reduces searchability, interoperability, and practical use [9,10,11,12,13].

These issues are especially relevant in ArcGIS Hub, where catalog search and item presentation depend on ArcGIS Online item details, categories, tags, sharing settings, and, when enabled, standards-based metadata. ArcGIS distinguishes between item information visible on the item page and fuller metadata records maintained through an organizational metadata style [14,15,16]. Consequently, an item may be publicly accessible and easy to open yet lack the technical documentation required for advanced geospatial analysis.

North Carolina Central University’s CoDE Open Data Hub is being developed as a university geospatial knowledge platform with support from NASA award 22-MUREPDEAP-0002 and National Science Foundation award 2226312. Part of this study measures the quality and discoverability of the catalog as it existed on 19 June 2026 and establishes a repeatable benchmark for subsequent development.

The study addresses three research questions:

• To what extent do current CoDE Hub items support Findability, Accessibility, Interoperability, and Reusability through the metadata evidence available in ArcGIS item records, formal metadata, and layer properties?
• How do FAIR-aligned and FGDC CSDGM-informed metadata scores vary across item types and individual criteria?
• How effectively can public users locate Hub items through exact-title, thematic, geographic, type-based, and category-based discovery pathways?

The study combines a complete catalog census, programmatic extraction of live ArcGIS metadata, a transparent 0–2 scoring framework, and a structured discoverability protocol. FAIR provides the organizing framework for interpretation, while FGDC CSDGM supplies the geospatial metadata elements used to construct measurable indicators.

2. Background and Conceptual Framework

2.1. University Geospatial Hubs as Spatial Data Infrastructure

University geospatial hubs occupy an intermediate position between institutional repositories, teaching environments, and public spatial data infrastructures. Unlike narrowly defined data catalogs, they commonly contain hosted feature layers, web maps, dashboards, StoryMaps, reports, images, instructional files, and project applications. This heterogeneity increases the range of audiences served but also complicates consistent description, governance, and discovery. The challenge is not simply to expose content online; it is to connect diverse resources to sufficient contextual and technical documentation so that they can be interpreted across levels of expertise [2,17,18].

2.2. Metadata Quality and Geospatial Reuse

Metadata quality is multidimensional. Completeness addresses whether expected elements are present; clarity and specificity address whether those elements are meaningful; provenance and lineage support evaluation of origin and processing; and accessibility determines whether metadata can be found and used. Bruce and Hillmann [6] conceptualized metadata quality as a continuum rather than a simple present-or-absent condition, while Ochoa and Duval [8] demonstrated that scalable assessment can combine automated measures with explicit quality rules. For open data portals, automated assessments have shown that weak descriptions, inaccessible resources, inconsistent formats, and incomplete licensing information remain common even in mature catalogs [4,9,10,13].

Geospatial metadata requires additional technical elements because the meaning and suitability of spatial data depend on location, coordinate reference, scale, geometry, time, attributes, and processing history. The FGDC Content Standard for Digital Geospatial Metadata (CSDGM), Version 2 (FGDC-STD-001-1998) [19], organizes these requirements into the following sections: identification, data quality, spatial data organization, spatial reference, entity and attribute, distribution, and metadata reference. Related standards, ISO 19157:2013 [20] and ISO 19115-1:2014 [21], formalize geographic data-quality and metadata concepts for contemporary spatial data infrastructures. Research on geospatial data quality further emphasizes that quality information must be structured and accessible in ways that support dynamic fitness-for-use decisions [22,23]. Although current federal practice also recognizes ISO metadata standards, CSDGM remains a useful operational framework for ArcGIS organizations that use the FGDC metadata style and for legacy or project-specific records that follow that structure.

The persistent cost of metadata authorship has also motivated automated and semi-automated approaches. These methods derive metadata from dataset structure, geometry, processing workflows, web resources, or publication services, but they generally retain a role for human review and item-specific contextualization [24,25,26,27,28,29]. Recent large-scale FAIR assessment work also demonstrates the value of a customized rubric, explicit applicability rules, calibration, and retained evidence when diverse data products are assessed consistently [30]. This distinction is central to the CoDE workflow: automation can improve coverage and consistency, but technical and interpretive fields still require accountable editorial judgment.

2.3. FAIR Principles as the Organizing Framework

The FAIR principles define four related goals for the stewardship of digital resources: they should be findable through rich metadata and searchable systems, accessible through clear retrieval mechanisms, interoperable through shared structures and representations, and reusable through sufficient contextual, licensing, provenance, and domain-specific information [5]. FAIR is intentionally general, so assessment requires indicators suited to the data community and technical environment. The Research Data Alliance FAIR Data Maturity Model therefore treats FAIR evaluation as a community-specific operationalization task in which indicators, priorities, applicability, and scoring procedures must be made explicit [31]. Recent analyses similarly show that FAIR assessment tools differ substantially in their metrics and that a metric may operationally assess a different or broader function than its declared FAIR dimension [30,32,33,34].

For geospatial resources, FAIR implementation depends on metadata that describe spatial reference, geometry, geographic coverage, attributes, distribution, provenance, and use conditions. Geospatial studies have likewise emphasized the role of standards and minimum information profiles in making spatial data and GIS products more interoperable and reusable [35,36]. In this study, FAIR provides the conceptual structure, while FGDC CSDGM provides the domain-specific content areas from which the 19 assessment criteria were derived. The resulting scores are FAIR-aligned measures of the evidence available in the Hub, not a formal certification against every FAIR sub-principle.

2.4. Discoverability as an Outcome of Metadata and Catalog Design

Discoverability concerns whether a user can identify and reach an appropriate resource through search or browsing. It is influenced by metadata content, catalog indexing, ranking, information architecture, and navigation burden. Studies of open data portals distinguish discoverability from mere technical availability: a resource can be public yet remain difficult to locate without an exact title or direct link [37]. Research on geospatial discovery systems identifies title, subject, place, format, spatial footprint, and faceted browsing as core retrieval elements [2], while usability experiments show that both metadata records and catalog interfaces can obstruct spatial data discovery and selection [38]. Recent geospatial search research further demonstrates that semantic enrichment, natural-language processing, and ranking methods can improve retrieval relevance, although such methods require evaluation against explicit relevance judgments and user tasks [39]. Recent geoportal usability studies likewise show that navigation, search-result presentation, user guidance, map interaction, and metadata descriptions jointly shape discovery outcomes [40,41]. Meaningful themes and place keywords, predictable titles, consistent categories, clear item types, visible ranking, and manageable click-depth therefore support exploratory discovery.

2.5. Study Contribution

This study evaluates a heterogeneous university ArcGIS Hub catalog through a combined FAIR and geospatial metadata framework. It brings together a complete catalog census, programmatic extraction of live ArcGIS item information, FGDC CSDGM-informed scoring, and a structured discoverability test. The contribution is a practical method for showing how broad FAIR dimensions are expressed through specific geospatial metadata fields and public catalog functions.

3. Materials and Methods

3.1. Study Design and Assessment Date

This study used a descriptive, cross-sectional case-study design combining a complete metadata census with a proportionally stratified discoverability assessment. A case-study design was appropriate because the objective was to characterize the operational state of a heterogeneous institutional catalog while preserving item-level evidence and platform context. The complete census avoided sampling error for metadata-quality estimates, whereas manual discoverability testing was applied to a stratified sample because search, browsing, and click-depth observations required direct interaction with the public interface.

The assessment reflects the state of the Hub on 19 June 2026, following routine metadata completion work undertaken during platform development. FAIR was used as an organizing and interpretive framework rather than as a claim of formal certification. FGDC CSDGM was selected as the domain-specific content basis because the Hub contains geospatial resources, ArcGIS supports standards-based metadata, and the standard provides observable elements concerning identification, spatial reference, entities and attributes, distribution, provenance, and metadata responsibility. The assessment therefore evaluates publicly retrievable evidence that supports FAIR-related functions; it does not validate every mandatory or conditional CSDGM element or every FAIR sub-principle.

The 19-criterion scoring rules and the original 50% and 75% performance bands were fixed before the final assessment run. The same rules-based script was applied to every item, and the evidence supporting each criterion score was retained. Alternative weighting, applicability-normalized scoring, and threshold analyses were added during revision as post hoc robustness checks applied to the archived item-level scores. They are not presented as preregistered analyses and are distinct from the independent inter-assessor reliability assessment described in Section 3.6. The overall workflow for the metadata and discoverability assessment is shown in Figure 1.

Figure 1. Workflow for the FAIR- and FGDC CSDGM-informed metadata and discoverability assessment.

Figure 1. Workflow for the FAIR- and FGDC CSDGM-informed metadata and discoverability assessment.

3.2. Catalog Inventory and Unit of Analysis

The unit of analysis was an individual ArcGIS item included in the public CoDE Hub catalog. The catalog was defined from ArcGIS Online (2026.R1, February 2026 release) and ArcGIS Hub (cloud release current on 19 June 2026; Esri, Redlands, CA, USA) groups configured as Hub content sources [14,15,16]. Item identifiers were retrieved programmatically, deduplicated across groups, and matched to current ArcGIS item records. The final census contained 747 public items.

The inventory workflow comprised:

• retrieval and deduplication of item identifiers from catalog groups;
• collection of current item-page properties, including title, summary, description, tags, categories, credits, terms of use, access level, owner, item type, and creation and modification dates;
• download and parsing of standards-based metadata XML where available;
• retrieval of feature-layer geometry, spatial reference, and field information where accessible; and
• classification of each item according to its primary public-facing function.

3.3. FAIR- and FGDC CSDGM-Informed Assessment Framework

The assessment used FAIR as the organizing framework and FGDC CSDGM, Version 2 (FGDC-STD-001-1998) [19], as the geospatial metadata standard from which observable criteria were derived. The seven CSDGM content areas were operationalized as 19 criteria and mapped to FAIR through a functional crosswalk. Because one metadata element can support more than one FAIR objective, the revised mapping distinguishes a primary FAIR dimension from secondary FAIR relationships. The primary dimension represents the criterion’s most direct function in this assessment and is used for numerical aggregation so that one observed score is not counted more than once. Secondary relationships acknowledge conceptual overlap and guide interpretation. For example, access constraints are assigned primarily to Accessibility because they describe retrieval conditions, while their effect on permissible downstream use is recorded as a secondary relationship to Reusability; use constraints are assigned primarily to Reusability and secondarily to Accessibility.

Primary assignments followed three rules: the criterion had to be observable from the available public evidence; the assignment had to correspond to the most immediate user or machine function supported; and the relationship had to remain traceable to a FAIR principle or domain-relevant metadata requirement. The mapping is therefore an operational model for this catalog rather than an assertion that the criteria are mutually exclusive or sufficient for complete FAIR assessment. Persistent identifiers, long-term metadata persistence after data removal, machine-actionable controlled vocabularies, and qualified links between related resources were outside the available evidence. Reported FAIR scores summarize the selected indicators and do not certify full FAIR compliance [30,31,32]. The functional crosswalk between the FGDC CSDGM-informed criteria and their primary and secondary FAIR relationships is presented in Table 1.

Table 1. Functional crosswalk of FGDC CSDGM-informed criteria to primary and secondary FAIR relationships.

Criterion	FGDC Content Area	Primary FAIR Dimension	Secondary Relationship	Functional Rationale
Title clarity	Identification information	Findable	Reusable	Supports discovery and later interpretation.
Abstract/description	Identification information	Findable	Reusable	Rich description supports search and fitness-for-use judgments.
Purpose	Identification information	Reusable	Findable	Intended use supports reuse; purpose terms may also aid discovery.
Theme keywords	Identification information	Findable	Interoperable	Indexing terms support retrieval; controlled terms can support semantic interoperability.
Place keywords	Identification information	Findable	Interoperable	Geographic terms support spatial discovery; controlled gazetteers can support interoperability.
Access constraints	Identification information	Accessible	Reusable	States retrieval conditions and affects permissible downstream use.
Use constraints	Identification information	Reusable	Accessible	Defines permissions or limitations and may include access-related conditions.
Data source	Data quality information	Reusable	Interoperable	Source attribution supports trust, provenance, and linking.
Lineage/process notes	Data quality information	Reusable	Interoperable	Processing history supports provenance and integration decisions.
Completeness/limitations	Data quality information	Reusable	None	Supports fitness-for-use and responsible interpretation.
Data type/geometry	Spatial data organization	Interoperable	Findable	Resource structure supports integration and type-based discovery.
Spatial reference	Spatial reference information	Interoperable	Reusable	Coordinate reference supports correct integration and fitness-for-use.
Attribute definitions	Entity and attribute information	Interoperable	Reusable	Schema meaning supports interpretation, integration, and reuse.
Access/download options	Distribution information	Accessible	Findable	Provides retrieval mechanisms and a reachable resource path.
Distribution format	Distribution information	Accessible	Interoperable	Describes delivery mechanism and technical compatibility.
Distributor/contact	Distribution information	Accessible	Reusable	Identifies a responsible access point and accountable steward.
Date created/published	Metadata reference information	Findable	Reusable	Supports indexing, citation context, and temporal interpretation.
Date modified/updated	Metadata reference information	Findable	Reusable	Supports discovery by currency and assessment of currentness.
Metadata contact/date	Metadata reference information	Reusable	Accessible	Supports metadata responsibility, maintenance, and clarification.

Table 1. Functional crosswalk of FGDC CSDGM-informed criteria to primary and secondary FAIR relationships.

Criterion	FGDC Content Area	Primary FAIR Dimension	Secondary Relationship	Functional Rationale
Title clarity	Identification information	Findable	Reusable	Supports discovery and later interpretation.
Abstract/description	Identification information	Findable	Reusable	Rich description supports search and fitness-for-use judgments.
Purpose	Identification information	Reusable	Findable	Intended use supports reuse; purpose terms may also aid discovery.
Theme keywords	Identification information	Findable	Interoperable	Indexing terms support retrieval; controlled terms can support semantic interoperability.
Place keywords	Identification information	Findable	Interoperable	Geographic terms support spatial discovery; controlled gazetteers can support interoperability.
Access constraints	Identification information	Accessible	Reusable	States retrieval conditions and affects permissible downstream use.
Use constraints	Identification information	Reusable	Accessible	Defines permissions or limitations and may include access-related conditions.
Data source	Data quality information	Reusable	Interoperable	Source attribution supports trust, provenance, and linking.
Lineage/process notes	Data quality information	Reusable	Interoperable	Processing history supports provenance and integration decisions.
Completeness/limitations	Data quality information	Reusable	None	Supports fitness-for-use and responsible interpretation.
Data type/geometry	Spatial data organization	Interoperable	Findable	Resource structure supports integration and type-based discovery.
Spatial reference	Spatial reference information	Interoperable	Reusable	Coordinate reference supports correct integration and fitness-for-use.
Attribute definitions	Entity and attribute information	Interoperable	Reusable	Schema meaning supports interpretation, integration, and reuse.
Access/download options	Distribution information	Accessible	Findable	Provides retrieval mechanisms and a reachable resource path.
Distribution format	Distribution information	Accessible	Interoperable	Describes delivery mechanism and technical compatibility.
Distributor/contact	Distribution information	Accessible	Reusable	Identifies a responsible access point and accountable steward.
Date created/published	Metadata reference information	Findable	Reusable	Supports indexing, citation context, and temporal interpretation.
Date modified/updated	Metadata reference information	Findable	Reusable	Supports discovery by currency and assessment of currentness.
Metadata contact/date	Metadata reference information	Reusable	Accessible	Supports metadata responsibility, maintenance, and clarification.

3.4. Scoring, Weighting, Applicability, and Descriptive Performance Bands

Each applicable criterion was scored on a three-level ordinal scale. A score of 0 indicated that no usable evidence was available; 1 indicated evidence that was present but generic, incomplete, vague, or insufficient for confident interpretation; and 2 indicated evidence that was specific, clear, and useful for the discovery, access, interoperability, or reuse function represented by the criterion. The three levels distinguish absence, nominal presence, and operational usefulness without implying measurement precision unsupported by the metadata. This approach also avoids treating a populated field as equivalent to a useful field. Criterion-specific rules were applied consistently across item-page properties, formal metadata XML, and accessible layer or schema properties, and the score and evidence string were retained for every item and criterion [6,8,10,12,30].

The evidence hierarchy prioritized item-specific information over generic or inherited text. Where the same criterion appeared in more than one source, explicit item-level metadata or accessible layer properties were used before organizational boilerplate. Generic statements could receive partial credit but could not receive the same score as specific, item-relevant evidence. The complete decision rules are shown in

Table 2. Criterion-specific operational definitions for scores of 0, 1, and 2.

Criterion	0—Missing	1—Partial/Weak	2—Complete/Useful
Title clarity	Blank or absent	Date/code-only, generic, or unclear title	Specific, understandable, distinguishable title
Abstract/description	No abstract or description	Present but short, generic, or placeholder-like	Specific description with sufficient explanatory detail
Purpose	No purpose statement	Present but short or generic	Clear intended purpose or use
Theme keywords	No meaningful theme keywords	One or two meaningful keywords	Three or more meaningful subject keywords
Place keywords	No geographic keyword or description	Only broad geography such as North Carolina	Specific place keywords or clear geographic coverage
Access constraints	Access status unavailable or item inaccessible	Public status only, without an explicit constraint statement	Public status plus explicit access-constraint information
Use constraints	No terms, license, or use limitation	Generic or brief statement	Clear restrictions, permissions, license, or item-relevant limitations
Data source	No source or credit	Only owner/publisher or generic organization	Specific source, creator, program, agency, or source URL
Lineage/process notes	No process information	Generic or brief process statement	Specific creation, processing, geocoding, derivation, or update information
Completeness/limitations	No limitations or quality information	Generic disclaimer	Clear accuracy, completeness, limitation, uncertainty, or known-issue statement
Data type/geometry	Type unavailable	General item type only	Specific resource type and/or geometry information
Spatial reference	No coordinate-system information	WKID or generic spatial representation only	Coordinate system, projection, datum, or equivalent reference sufficiently identified
Attribute definitions	No field or attribute information	Fields exist or a generic attribute statement is present	Useful definitions of fields, variables, units, codes, or domains
Access/download options	Item inaccessible	View-only access	Public view plus download, service, file, API, or open-resource access
Distribution format	No format information	Item type only	Specific file type, service type, or distribution method
Distributor/contact	No publisher or contact	Owner or organization only	Named person/unit, email, phone, or clearly responsible organization
Date created/published	No date	Only metadata date or unclear date	Creation/publication date available
Date modified/updated	No date	Only unclear revision reference	Modification/update date available
Metadata contact/date	Neither available	Contact or metadata date available	Both metadata contact and metadata date available

and reproduced in the archived scoring-rules file.

Equal criterion weighting was retained for the primary 38-point index because the study did not possess validated stakeholder-derived or empirically estimated importance weights. Imposing unequal weights would therefore add normative assumptions that could not be defended from the available evidence. Each criterion contributes at most two points, which makes the index transparent and reproducible. Because the 38-point total gives greater influence to FAIR dimensions represented by more criteria, an equal-dimension sensitivity analysis was also conducted by averaging the four normalized FAIR-dimension percentages, with each dimension contributing 25%.

As a post hoc sensitivity analysis, catalog heterogeneity was addressed through a core-plus-extension model. Seventeen criteria form a common profile for identification, discovery, access, provenance, use conditions, responsibility, and currency. Spatial reference and attribute definitions form a technical extension that is applicable to data-bearing services and files and is treated as not applicable to narrative or composite products without an exposed schema. Applicability-normalized percentages were calculated as observed points divided by the maximum points for the criteria applicable to an item. The common 19-criterion score is retained as a catalog-wide benchmark, while the type-aware score tests whether conclusions depend on structurally irrelevant criteria.

The 50% and 75% cut points were derived from the ordinal anchors. A score of 50% corresponds to an average of 1.0 per applicable criterion, the threshold for partial or weak evidence. A score of 75% corresponds to an average of 1.5, indicating performance midway between partial and complete evidence and therefore predominantly complete or useful. Scores below 50% average less than partial evidence. These categories are descriptive performance bands, not FGDC compliance classes or FAIR certification thresholds; continuous scores and criterion-level distributions remain the primary results.

For the common benchmark, the maximum total score was 38 and percentage scores were calculated as (observed score/38) × 100. The resulting percentage is a transparent descriptive index of the evidence observed across the 19 criteria; it is not a claim that the intervals between 0, 1, and 2 represent a continuous physical quantity.

FAIR-aligned dimension scores were calculated for each item by summing the criteria assigned primarily to each dimension and dividing by that dimension’s maximum possible score. Findable included six criteria (maximum 12), Accessible four (maximum 8), Interoperable three (maximum 6), and Reusable six (maximum 12). Mean percentages were then calculated across all 747 items. The original 38-point total and FGDC content-area summaries were retained so that FAIR interpretation did not replace the more detailed geospatial metadata results. Robustness was examined using equal FAIR-dimension weighting, applicability-normalized scores, and alternative 70/50 and 80/60 performance-band cut points. The descriptive performance bands used in the primary assessment are summarized in Table 3.

Table 3. Descriptive performance bands derived from the ordinal score anchors.

Performance Band	Percentage Range	Average Criterion Score	Interpretation
High	75.00–100.00%	1.50–2.00	Predominantly complete/useful evidence
Moderate	50.00–74.99%	1.00–1.49	At least partial/weak evidence on average
Low	Below 50.00%	Below 1.00	Less than partial/weak evidence on average

Table 3. Descriptive performance bands derived from the ordinal score anchors.

Performance Band	Percentage Range	Average Criterion Score	Interpretation
High	75.00–100.00%	1.50–2.00	Predominantly complete/useful evidence
Moderate	50.00–74.99%	1.00–1.49	At least partial/weak evidence on average
Low	Below 50.00%	Below 1.00	Less than partial/weak evidence on average

3.5. Discoverability Assessment

Discoverability was assessed for a proportionally stratified sample of 153 items, representing 20.48% of the catalog. Sampling was conducted independently within each broad item-type class using a fixed random seed of 20260619. The stratum sample size was the ceiling of 20% of the stratum, with a minimum of one item, ensuring representation of all eight broad classes: datasets/hosted feature layers (45 of 223), web maps (33 of 161), web applications (13 of 64), dashboards (3 of 12), StoryMaps (11 of 52), document/file resources (35 of 173), training resources (1 of 3), and other items (12 of 59). The protocol approximated the experience of a public user navigating the Hub without administrator privileges.

The five search and browsing tests were scored 0 (unsuccessful), 1 (partially successful), or 2 (successful) using the rules in

Table 4. Discoverability test protocol and criterion-specific scoring logic.

Test	Procedure	0/1/2 Scoring Logic	Purpose
Exact-title search	Search the complete public item title.	0: not found; 1: found only after minor title normalization or one refinement; 2: found with the complete title query.	Known-item retrieval
Keyword search	Search one or two substantive terms from the title or description.	0: not found; 1: found after one refinement or difficult to distinguish among broad results; 2: found with the initial substantive query.	Thematic retrieval without exact-title knowledge
Location search	Search a relevant place name when applicable.	0: not found; 1: returned among many same-context results or after one refinement; 2: clearly retrieved by a specific place query.	Place-based retrieval
Data-type search	Search or filter by broad item type, such as web map, dataset, dashboard, or StoryMap.	0: not found; 1: found only after refinement; 2: found through the initial type-based pathway.	Retrieval by resource type
Category browsing	Navigate from the homepage through catalog categories or collections.	0: no appropriate path; 1: indirect path or additional traversal required; 2: appropriate category directly retrieved the item.	Browsing through information architecture
Click-depth	Count discrete user-initiated actions from the homepage to the public item page or usable resource.	Recorded separately; scrolling and browser-back actions were not counted.	Navigation burden

. Location was excluded from an item’s denominator when no meaningful place query could be derived. Click-depth was recorded separately. A search-based percentage was computed from applicable tests, and the final discoverability band combined that percentage with navigation burden so that a high retrieval score did not automatically imply an efficient path. The archived assessment did not record exact numerical rank or first-page status; the reported results therefore measure retrieval success rather than rank-sensitive findability. A repeated assessment should retain target rank, first-page/top-k status, default sort, query text, and result count.The criteria used to assign the final discoverability performance bands are summarized in Table 5.

Table 5. Discoverability performance bands.

Band	Search Score	Click-Depth	Interpretation
High	At least 75%	No more than 3 clicks	Strong retrieval and a direct navigation path
Moderate	At least 50%	No more than 4 clicks	Adequate retrieval with limited additional navigation
Low	Below 50% or more than 4 clicks	More than 4 clicks where applicable	Weak retrieval or substantial navigation burden

Table 5. Discoverability performance bands.

Band	Search Score	Click-Depth	Interpretation
High	At least 75%	No more than 3 clicks	Strong retrieval and a direct navigation path
Moderate	At least 50%	No more than 4 clicks	Adequate retrieval with limited additional navigation
Low	Below 50% or more than 4 clicks	More than 4 clicks where applicable	Weak retrieval or substantial navigation burden

3.6. Analysis, Robustness, and Quality Control

Programmatic metadata extraction and analysis were conducted using ArcGIS Notebooks Runtime 13.0 (Esri, Redlands, CA, USA), within the ArcGIS Online 2026.R1 environment. The runtime included ArcGIS API for Python 2.4.2. The automated script retained the evidence used for every criterion score and generated a separate scoring-rules file, permitting item-level inspection and deterministic reruns. The rules and output columns were reviewed against the manuscript tables, and arithmetic totals were recalculated from the criterion-level score columns using a separate verification script. Independent inter-assessor reliability was evaluated by having a second assessor rescore a 50-item stratified subsample (6.69% of the 747-item catalog) using the frozen 19-criterion rubric and the same 0–2 decision rules. The subsample approximated the catalog distribution while retaining all eight broad item classes: 15 datasets/hosted feature layers, 11 web maps, 4 web applications, 1 dashboard, 3 StoryMaps, 11 document/file resources, 1 training resource, and 4 other items. Reliability was calculated across 950 paired criterion ratings using exact agreement and linear weighted Cohen’s kappa, with quadratic weighted kappa as a sensitivity statistic. Agreement between the two 38-point item totals was evaluated with a two-way random-effects, absolute-agreement, single-measure intraclass correlation coefficient (ICC[A,1]). Performance-band agreement was summarized using percent agreement and Cohen’s kappa. The second-assessor ratings were used to quantify reproducibility and did not replace the original census scores.

The metadata analysis used descriptive statistics because the assessment covered the complete catalog census. Summary outputs included item counts by type; mean, median, minimum, and maximum scores; performance-band frequencies; criterion and dimension completeness; and mean scores by item type. Robustness analyses compared the equal-criterion benchmark with equal FAIR-dimension weighting, an applicability-normalized core-plus-extension score, and alternative classification thresholds. Spearman rank correlation and performance-band agreement were used to assess whether conclusions changed under these design choices. Discoverability outputs included success distributions for each test, final discoverability bands, and click-depth statistics.

4. Results

4.1. Catalog Composition

The final assessment included 747 public catalog items. Datasets or hosted feature layers were the largest group (223 items; 29.85%), followed by document or file resources (173; 23.16%) and web maps (161; 21.55%). Training resources were the least frequent category (3; 0.40%); the complete catalog composition by item type is reported in Table 6.

Table 6. Audited CoDE Hub items by type.

Percentage	Number	Item Type
29.85%	223	Dataset/hosted feature layer
21.55%	161	Web map
8.57%	64	Web application
1.61%	12	Dashboard
6.96%	52	StoryMap
23.16%	173	Document/file resource
0.40%	3	Training resource
7.90%	59	Other
100.00%	747	Total

Table 6. Audited CoDE Hub items by type.

Percentage	Number	Item Type
29.85%	223	Dataset/hosted feature layer
21.55%	161	Web map
8.57%	64	Web application
1.61%	12	Dashboard
6.96%	52	StoryMap
23.16%	173	Document/file resource
0.40%	3	Training resource
7.90%	59	Other
100.00%	747	Total

4.2. Overall Metadata Descriptive Performance

The mean metadata score was 28.57 of 38 (75.19%), and the median was 31. Scores ranged from 11 to 38. Using the predefined descriptive thresholds, 490 items (65.60%) were in the high performance band, 207 (27.71%) were moderate, and 50 (6.69%) were low (Table 7). The distribution of metadata quality percentage scores across the 747 items is shown in Figure 2.These bands summarize the observed evidence and should not be interpreted as FGDC compliance levels or FAIR certification.

Table 7. Overall metadata descriptive performance bands.

Percentage	Number	Performance Band
65.60%	490	High
27.71%	207	Moderate
6.69%	50	Low
100.00%	747	Total

Table 7. Overall metadata descriptive performance bands.

Percentage	Number	Performance Band
65.60%	490	High
27.71%	207	Moderate
6.69%	50	Low
100.00%	747	Total

Figure 2. Distribution of metadata quality percentage scores across the 747 items.

Figure 2. Distribution of metadata quality percentage scores across the 747 items.

4.3. FAIR-Aligned Performance

The FAIR-aligned results show the strongest performance for Findability (87.25%) and Accessibility (85.96%). Reusability averaged 67.26%, while Interoperability was lowest at 52.59%. The pattern reflects strong descriptive, indexing, distribution, and access information across the catalog, alongside weaker spatial-reference and attribute documentation as shown in Table 8.

Table 8. Mean FAIR-aligned metadata scores across 747 items.

Indicators Included	Mean %	Mean Score/Max	Criteria (n)	FAIR Dimension
Title, abstract, theme/place keywords, creation and modification dates	87.25%	10.47/12	6	Findable
Access constraints, access/download, distribution format, distributor/contact	85.96%	6.88/8	4	Accessible
Data type/geometry, spatial reference, attribute definitions	52.59%	3.16/6	3	Interoperable
Purpose, use constraints, source, lineage, limitations, metadata contact/date	67.26%	8.07/12	6	Reusable

Table 8. Mean FAIR-aligned metadata scores across 747 items.

Indicators Included	Mean %	Mean Score/Max	Criteria (n)	FAIR Dimension
Title, abstract, theme/place keywords, creation and modification dates	87.25%	10.47/12	6	Findable
Access constraints, access/download, distribution format, distributor/contact	85.96%	6.88/8	4	Accessible
Data type/geometry, spatial reference, attribute definitions	52.59%	3.16/6	3	Interoperable
Purpose, use constraints, source, lineage, limitations, metadata contact/date	67.26%	8.07/12	6	Reusable

Findability was supported by strong titles, abstracts, theme keywords, and complete item dates. Accessibility benefited from clear distribution formats and identifiable contacts. The lower Interoperability score was driven primarily by incomplete spatial-reference and attribute documentation. Reusability was limited by uneven lineage and limitation statements, although purpose, use constraints, and metadata responsibility were generally stronger.

4.4. Completeness by FGDC-Informed Dimension and Criterion

Performance varied substantially across dimensions. Spatial data organization was strongest (98.26%), followed by metadata reference information (91.70%) and distribution information (90.90%). Identification information averaged 79.87%. Table 9 showed that data quality information was lower (54.66%), and the weakest dimensions were entity and attribute information (43.51%) and spatial reference information (16.00%).

Table 9. Metadata completeness by FGDC-informed dimension.

Weakest Criterion	Strongest Criterion	Mean %	Mean Score/Max	Dimension
Place keywords	Abstract/description	79.87%	11.18/14	Identification information
Lineage/process notes	Data source	54.66%	3.28/6	Data quality information
Data type/geometry	Data type/geometry	98.26%	1.97/2	Spatial data organization
Spatial reference	Spatial reference	16.00%	0.32/2	Spatial reference information
Attribute definitions	Attribute definitions	43.51%	0.87/2	Entity and attribute information
Access/download options	Distribution format	90.90%	5.45/6	Distribution information
Metadata contact/date	Date created/published	91.70%	5.50/6	Metadata reference information

Table 9. Metadata completeness by FGDC-informed dimension.

Weakest Criterion	Strongest Criterion	Mean %	Mean Score/Max	Dimension
Place keywords	Abstract/description	79.87%	11.18/14	Identification information
Lineage/process notes	Data source	54.66%	3.28/6	Data quality information
Data type/geometry	Data type/geometry	98.26%	1.97/2	Spatial data organization
Spatial reference	Spatial reference	16.00%	0.32/2	Spatial reference information
Attribute definitions	Attribute definitions	43.51%	0.87/2	Entity and attribute information
Access/download options	Distribution format	90.90%	5.45/6	Distribution information
Metadata contact/date	Date created/published	91.70%	5.50/6	Metadata reference information

At criterion level, creation and modification dates were complete for all items. Data type or geometry was complete for 96.52%, abstracts or descriptions for 95.72%, and distribution format for 95.85%. The largest technical gap was spatial reference: 71.89% of items lacked coordinate-system evidence and only 3.88% received a complete score. Attribute definitions were complete for only 0.80%, although 85.41% received partial credit because fields or generic attribute statements were present. Place keywords were complete for 10.98%; lineage was complete for 25.17%; and completeness or limitation statements were complete for 25.84%. Table 10 below shows these results. The distribution of complete, partial, and missing metadata across individual criteria is illustrated in Figure 3.

Table 10. Metadata completeness by individual criterion.

Mean/2	Missing n (%)	Partial n (%)	Complete n (%)	Criterion
1.75	0 (0.00%)	187 (25.03%)	560 (74.97%)	Title clarity
1.95	6 (0.80%)	26 (3.48%)	715 (95.72%)	Abstract/description
1.76	12 (1.61%)	153 (20.48%)	582 (77.91%)	Purpose
1.85	6 (0.80%)	97 (12.99%)	644 (86.21%)	Theme keywords
0.92	144 (19.28%)	521 (69.75%)	82 (10.98%)	Place keywords
1.42	133 (17.80%)	165 (22.09%)	449 (60.11%)	Access constraints
1.53	174 (23.29%)	6 (0.80%)	567 (75.90%)	Use constraints
1.28	160 (21.42%)	220 (29.45%)	367 (49.13%)	Data source
0.98	205 (27.44%)	354 (47.39%)	188 (25.17%)	Lineage/process notes
1.03	174 (23.29%)	380 (50.87%)	193 (25.84%)	Completeness/limitations
1.97	0 (0.00%)	26 (3.48%)	721 (96.52%)	Data type/geometry
0.32	537 (71.89%)	181 (24.23%)	29 (3.88%)	Spatial reference
0.87	103 (13.79%)	638 (85.41%)	6 (0.80%)	Attribute definitions
1.61	133 (17.80%)	23 (3.08%)	591 (79.12%)	Access/download options
1.96	0 (0.00%)	31 (4.15%)	716 (95.85%)	Distribution format
1.88	0 (0.00%)	88 (11.78%)	659 (88.22%)	Distributor/contact
2.00	0 (0.00%)	0 (0.00%)	747 (100.00%)	Date created/published
2.00	0 (0.00%)	0 (0.00%)	747 (100.00%)	Date modified/updated
1.50	176 (23.56%)	20 (2.68%)	551 (73.76%)	Metadata contact/date

Table 10. Metadata completeness by individual criterion.

Mean/2	Missing n (%)	Partial n (%)	Complete n (%)	Criterion
1.75	0 (0.00%)	187 (25.03%)	560 (74.97%)	Title clarity
1.95	6 (0.80%)	26 (3.48%)	715 (95.72%)	Abstract/description
1.76	12 (1.61%)	153 (20.48%)	582 (77.91%)	Purpose
1.85	6 (0.80%)	97 (12.99%)	644 (86.21%)	Theme keywords
0.92	144 (19.28%)	521 (69.75%)	82 (10.98%)	Place keywords
1.42	133 (17.80%)	165 (22.09%)	449 (60.11%)	Access constraints
1.53	174 (23.29%)	6 (0.80%)	567 (75.90%)	Use constraints
1.28	160 (21.42%)	220 (29.45%)	367 (49.13%)	Data source
0.98	205 (27.44%)	354 (47.39%)	188 (25.17%)	Lineage/process notes
1.03	174 (23.29%)	380 (50.87%)	193 (25.84%)	Completeness/limitations
1.97	0 (0.00%)	26 (3.48%)	721 (96.52%)	Data type/geometry
0.32	537 (71.89%)	181 (24.23%)	29 (3.88%)	Spatial reference
0.87	103 (13.79%)	638 (85.41%)	6 (0.80%)	Attribute definitions
1.61	133 (17.80%)	23 (3.08%)	591 (79.12%)	Access/download options
1.96	0 (0.00%)	31 (4.15%)	716 (95.85%)	Distribution format
1.88	0 (0.00%)	88 (11.78%)	659 (88.22%)	Distributor/contact
2.00	0 (0.00%)	0 (0.00%)	747 (100.00%)	Date created/published
2.00	0 (0.00%)	0 (0.00%)	747 (100.00%)	Date modified/updated
1.50	176 (23.56%)	20 (2.68%)	551 (73.76%)	Metadata contact/date

Figure 3. Complete, partial, and missing metadata by criterion.

Figure 3. Complete, partial, and missing metadata by criterion.

4.5. Metadata Descriptive Performance by Item Type

Benchmark metadata performance differed by item type. Dashboards had the highest mean percentage (82.46%), followed by document or file resources (82.31%) and StoryMaps (81.88%). Datasets or hosted feature layers had the lowest mean percentage (66.10%), and items in the residual other category averaged 70.29%. The lower benchmark score for datasets reflects both genuine technical documentation gaps and the greater relevance of spatial reference, attribute, lineage, and limitation information to data-bearing products; the type-aware sensitivity analysis in Section 4.7 addresses this applicability issue. Benchmark metadata performance by item type is summarized in Table 11.

Table 11. Benchmark metadata descriptive performance by item type.

Performance Band	Mean %	Mean Score/38	n	Item Type
Moderate	66.10%	25.12	223	Dataset/hosted feature layer
High	78.62%	29.88	161	Web map
High	76.69%	29.14	64	Web application
High	82.46%	31.33	12	Dashboard
High	81.88%	31.12	52	StoryMap
High	82.31%	31.28	173	Document/file resource
High	76.31%	29.00	3	Training resource
Moderate	70.29%	26.71	59	Other

Table 11. Benchmark metadata descriptive performance by item type.

Performance Band	Mean %	Mean Score/38	n	Item Type
Moderate	66.10%	25.12	223	Dataset/hosted feature layer
High	78.62%	29.88	161	Web map
High	76.69%	29.14	64	Web application
High	82.46%	31.33	12	Dashboard
High	81.88%	31.12	52	StoryMap
High	82.31%	31.28	173	Document/file resource
High	76.31%	29.00	3	Training resource
Moderate	70.29%	26.71	59	Other

4.6. Discoverability

The discoverability assessment included 153 items and represented every broad catalog class. The mean search-based discoverability percentage was 80.00%, and the mean click depth was 4.07, with a range of 1 to 7 clicks. After incorporating navigation burden, 43 items (28.10%) were classified as highly discoverable, 58 (37.91%) as moderately discoverable, and 52 (33.99%) as low discoverability. Because exact result rank was not archived, these bands describe retrieval success and navigation burden rather than first-page or top-k visibility (Table 12).

Table 12. Discoverability classification of sampled items.

Percentage	Number	Discoverability Class
28.10%	43	High
37.91%	58	Moderate
33.99%	52	Low
100.00%	153	Total

Table 12. Discoverability classification of sampled items.

Percentage	Number	Discoverability Class
28.10%	43	High
37.91%	58	Moderate
33.99%	52	Low
100.00%	153	Total

Exact-title and data-type searches were successful for every sampled item. Keyword search was successful for 88.24% and partially successful for the remaining 11.76%. Location search was fully successful for 4.03% of applicable items and partially successful for 94.63%. Because the catalog is heavily concentrated in North Carolina, the location results mainly reflect limited differentiation among resources that share the same broad geographic context. Category browsing was fully successful for 29.41%, partially successful for 48.37%, and unsuccessful for 22.22%. Exact-title, keyword, and item-type searches, therefore, performed strongly, while category structure and navigation depth were the more consequential constraints on exploratory use (Table 13). The distribution of click depth for the discoverability sample is shown in Figure 4.

Table 13. Search and browsing performance.

Unsuccessful n (%)	Partial n (%)	Successful n (%)	Test
0 (0.00%)	0 (0.00%)	153 (100.00%)	Exact-title search
0 (0.00%)	18 (11.76%)	135 (88.24%)	Keyword search
2 (1.34%)	141 (94.63%)	6 (4.03%)	Location search
0 (0.00%)	0 (0.00%)	153 (100.00%)	Data-type search
34 (22.22%)	74 (48.37%)	45 (29.41%)	Category browsing

Table 13. Search and browsing performance.

Unsuccessful n (%)	Partial n (%)	Successful n (%)	Test
0 (0.00%)	0 (0.00%)	153 (100.00%)	Exact-title search
0 (0.00%)	18 (11.76%)	135 (88.24%)	Keyword search
2 (1.34%)	141 (94.63%)	6 (4.03%)	Location search
0 (0.00%)	0 (0.00%)	153 (100.00%)	Data-type search
34 (22.22%)	74 (48.37%)	45 (29.41%)	Category browsing

Figure 4. Click-depth distribution for the discoverability sample.

Figure 4. Click-depth distribution for the discoverability sample.

4.7. Robustness and Sensitivity Analyses

Under equal FAIR-dimension weighting, the catalog mean was 73.26%, compared with 75.19% under equal criterion weighting. The two scores were strongly associated (Spearman rho = 0.988), and 716 of 747 items (95.85%) retained the same descriptive performance band. Equal-dimension weighting classified 467 items (62.52%) as high, 234 (31.33%) as moderate, and 46 (6.16%) as low. The broad pattern of stronger Findability and Accessibility and weaker Interoperability and Reusability was unchanged.

In the core-plus-extension sensitivity analysis, spatial reference and attribute definitions were treated as applicable to datasets or hosted feature layers and excluded from the denominator for other broad item types without exposed schema properties. The applicability-normalized mean was 79.93%. Item rankings remained highly consistent with the common benchmark (Spearman rho = 0.977), and 697 items (93.31%) retained the same performance band. This indicates that the common index is useful as a catalog benchmark, while type-aware normalization provides a more defensible comparison across heterogeneous products.

Performance-band proportions were more sensitive to the cut points than item rankings. Using 70% and 50% as cut points produced 551 high (73.76%), 146 moderate (19.54%), and 50 low items (6.69%). Using stricter 80% and 60% cut points produced 404 high (54.08%), 183 moderate (24.50%), and 160 low items (21.42%). Continuous scores, criterion distributions, and dimension profiles are therefore emphasized over the categorical labels.

The weighting, applicability, and threshold-sensitivity results are consolidated in

Table 14. Robustness and sensitivity analysis of weighting, applicability, and performance-band thresholds.

Scenario	Mean (%)	High n (%)	Moderate n (%)	Low n (%)	Stability Relative to Benchmark
Equal-criterion benchmark (19 criteria; 75/50 bands)	75.19	490 (65.60%)	207 (27.71%)	50 (6.69%)	Reference
Equal FAIR-dimension weighting (75/50 bands)	73.26	467 (62.52%)	234 (31.33%)	46 (6.16%)	rho = 0.988; 716 (95.85%) same band
Core-plus-extension applicability normalization (75/50 bands)	79.93	529 (70.82%)	179 (23.96%)	39 (5.22%)	rho = 0.977; 697 (93.31%) same band
Equal-criterion score with 70/50 cut points	75.19	551 (73.76%)	146 (19.54%)	50 (6.69%)	686 (91.83%) same band; ranking unchanged
Equal-criterion score with 80/60 cut points	75.19	404 (54.08%)	183 (24.50%)	160 (21.42%)	551 (73.76%) same band; ranking unchanged

Note: Spearman rho and same-band agreement compare each alternative scoring method with the equal-criterion benchmark using 75% and 50% cut points. Threshold-only scenarios retain the same continuous scores and rankings; therefore, Spearman rho is not repeated for those rows.

.

4.8. Inter-Assessor Reliability

Independent rescoring of 50 items produced exact agreement for 729 of 950 criterion-level ratings (76.74%). Linear weighted Cohen’s kappa was 0.587, and quadratic weighted kappa was 0.630. The original scores for this subsample averaged 28.86 of 38 (75.95%), compared with 29.60 of 38 (77.89%) for the second assessor. The absolute-agreement ICC(A,1) for total scores was 0.715. Performance-band classifications agreed for 41 of 50 items (82.00%), with a band-level Cohen’s kappa of 0.566. Among the 221 disagreements, 189 (19.89% of all paired ratings) differed by one score level and 32 (3.37%) differed by two levels; the overall inter-assessor reliability results are summarized in

Table 15. Inter-assessor reliability results for the 50-item validation subsample.

Reliability Measure	Result
Items independently rescored	50 (6.69% of catalog)
Paired criterion ratings	950
Exact criterion agreement	76.74% (729/950)
Linear weighted Cohen’s kappa	0.587
Quadratic weighted Cohen’s kappa	0.630
Mean total score /38	Original: 28.86; second assessor: 29.60
Absolute-agreement ICC(A,1)	0.715
Performance-band agreement	82.00% (41/50)
Band-level Cohen’s kappa	0.566

Note: ICC(A,1) denotes a two-way random-effects, absolute-agreement, single-measure intraclass correlation coefficient.

.

Criterion-level agreement was strongest for data source (linear weighted kappa = 0.828), purpose (0.805), distributor/contact (0.706), and metadata contact/date (0.662). Date created/published and date modified/updated each had 100% exact agreement, but kappa was not estimable because both assessors assigned the same score to every sampled item. The lowest criterion-level kappas were observed for completeness/limitations (−0.018), access/download options (0.042), place keywords (0.145), and access constraints (0.161). These results indicate that the aggregate scoring framework is reproducible while also identifying interpretive criteria that would benefit from more explicit decision examples.

5. Discussion

5.1. FAIR Profile of the CoDE Hub

The assessment indicates a strong foundation for Findability (87.25%) and Accessibility (85.96%). Most items contain useful descriptions, theme keywords, clear item types, distribution information, responsible contacts, and current creation and modification dates. These features support known-item retrieval and routine public access. The 75.19% benchmark and the high-band share of 65.60% should, however, be interpreted as descriptive summaries of the selected evidence rather than certification. The equal-dimension and threshold analyses show that the overall pattern is stable even though the number of items assigned to a categorical band depends on the chosen cut points.

The value of the combined framework is not the single catalog percentage alone. By preserving criterion-level evidence from item information, formal metadata XML, and layer properties, the assessment links broad FAIR-related outcomes to specific fields that can be corrected. This addresses a recurring challenge in FAIR operationalization: assessment metrics necessarily embody implementation choices, and their declared FAIR dimension may not capture every function they evaluate [30,32,33,34]. The primary-secondary crosswalk and sensitivity analyses make those choices explicit.

5.2. Interoperability and Reusability

Interoperability was the weakest FAIR-aligned dimension (52.59%), while Reusability averaged 67.26%. Spatial-reference documentation was missing for most items, complete attribute definitions were rare, and lineage and limitation statements were often generic or incomplete. These gaps matter because coordinate systems, field definitions, units, codes, domains, processing history, and known limitations determine whether data can be integrated correctly and evaluated for fitness for purpose [22,23,38,42].

The contrast between strong Accessibility and weaker Interoperability and Reusability illustrates a broader governance sequence: institutions can publish and expose resources faster than they can document schemas, spatial reference, provenance, quality, and limitations. A portal can therefore be operationally open but analytically under-documented. Opening a resource answers whether it can be reached; technical metadata answer whether it can be integrated, interpreted, and reused responsibly. Similar differences between availability and reuse readiness have been reported across open-data and FAIR assessment contexts [4,9,10,12,13,30].

5.3. Discoverability and Catalog Navigation

The discoverability results distinguish known-item retrieval from exploratory navigation. Exact-title, keyword, and item-type searches performed strongly, while place-based search, category browsing, and click-depth exposed greater difficulty for users without prior knowledge of a specific item. The low full-success rate for location search aligns with weak place-keyword performance, although the measure has limited diagnostic value in a catalog where most resources share a North Carolina context. These results reinforce the distinction between metadata-based retrieval and the organization of the catalog interface [2,38].

The average click depth of 4.07 qualifies the strong search score: a relevant item may be retrievable but still require several navigation actions before the user reaches the item page or usable resource. Discoverability is therefore a system outcome jointly produced by metadata, indexing, result ranking, category architecture, and interface design [37,40,41]. Because the archived protocol did not record numerical rank, the current results should not be interpreted as top-k performance. Future evaluations should log first-page status and rank and may test semantic enrichment or NLP-based ranking, but these methods should be evaluated against transparent relevance judgments and user tasks rather than assumed to improve discovery automatically [39,41].

5.4. Implications for Ongoing CoDE Hub Development

The findings identify clear priorities for the next development cycle: item-specific spatial reference, field-level attribute documentation, processing lineage, explicit limitations, controlled place terms, and more consistent item-type requirements. These elements should be embedded in the publication lifecycle rather than addressed only through periodic cleanup. A sustainable workflow should assign responsibility before publication, require item-type templates, validate machine-retrievable fields, flag generic or missing evidence, and route exceptions for accountable review. Periodic audits can then measure drift and improvement. Prior research shows that metadata profiles, training, integrated authoring workflows, programmatic extraction, and recurring FAIR monitoring can improve consistency while reducing the burden on individual data producers [25,27,28,29,30,33,43,44].

The item-type results show why one undifferentiated checklist is inadequate for a catalog containing datasets, services, maps, applications, dashboards, StoryMaps, documents, and training materials. A common core is necessary for consistent identification, source attribution, access, use conditions, responsibility, and currency. Data-bearing products additionally require coordinate reference, schema, units, domains, processing history, and quality information. The strong correlation between the common benchmark and the applicability-normalized score indicates that the main conclusions are not an artifact of type handling, while the higher type-aware mean shows that structurally irrelevant criteria can affect comparisons among product classes.

Place-based discovery requires attention at both the metadata and interface levels. State-level tags are sufficient for some resources, while county, municipality, watershed, campus, or project-area terms are more useful for local data. Controlled geographic vocabularies, consistent place tags, coherent categories, direct catalog entry points, and reduced click-depth would make the current content structure more legible to users who do not already know item titles.

5.5. Methodological Contribution and Transferability

The methodological contribution is an evidence-preserving governance instrument rather than the 75.19% value itself. The workflow integrates item-page information, standards-based XML, and layer properties; operationalizes broad FAIR concepts with domain-specific metadata criteria; distinguishes primary from secondary FAIR relationships; retains human-readable evidence; and tests sensitivity to weighting, thresholds, and item-type applicability. Each missing or partial score points to a specific field, template, validation rule, or interface intervention. This complements metadata-automation and FAIR-monitoring research by using computational extraction to support accountable review rather than treating automation as a substitute for judgment [23,24,26,29,30,32,33,34,39,45].

The independent rescoring adds an empirical check on the repeatability of the rubric. Agreement was strongest for criteria supported by relatively explicit evidence, such as purpose, source attribution, and responsible contact information, while criteria requiring more interpretive judgments showed lower kappa values. This pattern supports retaining evidence strings and criterion-specific rules and suggests that future versions of the rubric should add worked examples for limitations, access/download status, place specificity, and access constraints.

The method is transferable to other university, local government, or community ArcGIS Hub catalogs, but the NCCU prevalence estimates are not statistically generalizable to all portals. Transfer requires adaptation of the item-type classes, organizational metadata style, controlled vocabularies, evidence hierarchy, applicability matrix, and scoring rules. FAIR supplies a common interpretive structure, while FGDC, ISO, or other community standards can supply locally appropriate measurable fields. Comparative multi-hub research is needed to determine which observed gaps are platform-specific and which represent broader metadata-governance patterns.

5.6. Limitations and Future Evaluation

Several limitations bound interpretation. First, the assessment represents one catalog state after metadata remediation and does not estimate the pre-intervention condition or causal effect of the remediation. Second, the authors participated in Hub development, creating potential self-assessment bias. Independent rescoring of 50 items provides an external check on the rubric, but the validation sample covers 6.69% of the catalog and criterion-level agreement was uneven; the full 747-item census was not independently duplicated. Third, the rubric evaluates publicly retrievable evidence and may classify undocumented information as missing even when it exists elsewhere; conversely, generic boilerplate can receive partial credit without being fully item-specific. Fourth, applicability and evidence availability differ among item types, so the common index is supplemented by a type-aware sensitivity analysis. Fifth, the FAIR mapping remains an operational interpretation and does not test persistent identifiers, metadata persistence after removal, machine-actionable vocabularies, qualified links, or every FAIR sub-principle. Sixth, search rank, indexing, categories, and interface paths are temporally unstable. Finally, one institutional case cannot establish prevalence across university hubs.

Future reliability work should expand independent rescoring beyond the 50-item validation sample and refine the decision rules for criteria with lower agreement, particularly completeness/limitations, access/download options, place keywords, and access constraints. Repeated assessments should measure temporal change, while task-based usability testing with students, researchers, agency partners, and community users should test whether formal metadata performance translates into efficient discovery and appropriate reuse. Comparative analysis of other university ArcGIS Hub catalogs would distinguish platform-specific findings from broader governance challenges.

6. Conclusions

This study established a current-state benchmark for metadata descriptive performance and discoverability in the NCCU CoDE Open Data Hub. Across 747 items, the equal-criterion benchmark was 75.19%, with 65.60% of items in the high descriptive performance band. Equal FAIR-dimension weighting produced a similar mean of 73.26% and preserved the performance band of 95.85% of items. Findability (87.25%) and Accessibility (85.96%) were strongest, Reusability averaged 67.26%, and Interoperability was lowest at 52.59%. Independent rescoring of 50 items yielded 76.74% exact criterion-level agreement, a linear weighted Cohen’s kappa of 0.587, and an absolute-agreement ICC of 0.715 for total scores. These values summarize the selected evidence and do not constitute formal FGDC validation or FAIR certification.

The Hub supports reliable known-item retrieval and routine public access but is less consistent for exploratory navigation and technical reuse. The broader contribution is a transparent, evidence-preserving method that combines FAIR interpretation with geospatial metadata standards, separates common from type-specific requirements, and converts audit findings into actionable governance controls. The method can be adapted to other heterogeneous hubs, while the case findings require comparative and longitudinal research before broader generalization.